between logic and life

Portrait of Molly Liu

When I'm not building things, you'll probably find me…

#RunningWithNoPlan

#RockNRollLover

#CuriousAboutEverything

#ProudFeminist

#LivingThroughExperience

Hiiiiiii everyone,

I'm Molly (Meihan) Liu — a software engineer based in Toronto.

My work lives between disciplines. I studied both computer science and biology, and that mix still shapes how I think: I'm drawn to the places where systems, people, and data meet. Today my practice spans computer vision, GPU computing, and full-stack web — from deploying deep-learning models on surgical robots, to distributing billion-scale vector indices across GPUs, to building real-time, event-driven web platforms.

I care about software that is precise under the hood and effortless on the surface — the kind of engineering you feel but don't notice. My goal is to build the mediums through which people and machines meet, and to make those connections feel a little more human.

→ View my CV

projects

selected work — software, systems & a little chemistry

  1. 01

    Multi-GPU Approximate Nearest Neighbor Search

    GPU Computing · Distributed Systems · Python · CUDA

    A multi-GPU ANNS system built on NVIDIA cuVS (CAGRA) and CuPy, distributing billion-scale vector indices across GPUs. Benchmarked four data-partitioning strategies — including K-Means sharding — on build time, throughput (QPS), and recall@k over the SIFT1B dataset.

  2. 02

    AI Meeting Assistant Platform

    Full-Stack · Event-Driven · Next.js 15 · TypeScript

    A real-time AI video-conferencing platform embedding conversational agents into live calls via the Stream Video SDK and OpenAI Realtime API. An event-driven post-call pipeline on Inngest resolves speakers and generates GPT-4o summaries, over a fully type-safe tRPC + DrizzleORM data layer.

  3. 03

    FindLocal — Local Business Discovery Platform

    Full-Stack Web · Serverless · REST API · AWS

    A full-stack discovery platform on Next.js 15 with interest-based recommendation ranking and location-aware search. Serverless backend on AWS DynamoDB & S3, a 12+ route REST API, Google OAuth, and a community review system validated with Jest & React Testing Library.

  4. 04

    Multi-Agent Travel Itinerary Planner

    Multi-Agent Systems · Async Python · LLM Orchestration

    A 6-agent travel planner orchestrated as a Pipeline with parallel search and a bounded budget-adjustment loop. Independent agents run concurrently via asyncio.gather; the full pipeline is re-implemented in Java (Spring Boot) and Go (Gin) on each language's native concurrency model.

  5. 05

    Smart Clinical Trial Search 🏆 Best Use of Google Cloud, PharmaHacks 2022

    Cheminformatics · R Shiny · Python · RDKit

    The first clinical-trial search engine driven by SMILES molecular-fingerprint similarity. A Python pipeline merges ClinicalTrials.gov and DrugBank data; RDKit Tanimoto similarity ranks trials on structurally similar compounds, served through an interactive R Shiny app.

skills

Languages
C/C++, Python, Java, Go, TypeScript / JavaScript, R, SQL, CUDA, Dart, HTML / CSS, Shell
Frameworks & Libraries
Next.js, React / React Native, tRPC, DrizzleORM, TanStack Query, Tailwind CSS, Node.js, Inngest, Stream SDK, FastAPI, Streamlit, Spring Boot, Gin, Flutter, R Shiny
Databases & Cloud
PostgreSQL, MySQL, AWS (DynamoDB, S3), Vercel, Google Cloud, Docker
AI / ML & Data
OpenAI API, NVIDIA cuVS (CAGRA), CuPy, OpenCV, YOLO, Ultralytics, PyTorch, pandas, NumPy, RDKit
Tools & Testing
Git, GitHub Actions, CI/CD, REST APIs, OpenAPI / Swagger, Postman, NGINX, PM2, Jest, RTL, pytest, Zod, Pydantic